CN112883651B - System-level testability design multi-objective optimization method based on improved PBI method - Google Patents

System-level testability design multi-objective optimization method based on improved PBI method Download PDF

Info

Publication number
CN112883651B
CN112883651B CN202110249020.2A CN202110249020A CN112883651B CN 112883651 B CN112883651 B CN 112883651B CN 202110249020 A CN202110249020 A CN 202110249020A CN 112883651 B CN112883651 B CN 112883651B
Authority
CN
China
Prior art keywords
population
individuals
individual
value
pbi
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Expired - Fee Related
Application number
CN202110249020.2A
Other languages
Chinese (zh)
Other versions
CN112883651A (en
Inventor
杨成林
高亮亮
鲜航
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Electronic Science and Technology of China
Original Assignee
University of Electronic Science and Technology of China
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by University of Electronic Science and Technology of China filed Critical University of Electronic Science and Technology of China
Priority to CN202110249020.2A priority Critical patent/CN112883651B/en
Publication of CN112883651A publication Critical patent/CN112883651A/en
Application granted granted Critical
Publication of CN112883651B publication Critical patent/CN112883651B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/12—Computing arrangements based on biological models using genetic models
    • G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Medical Informatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physiology (AREA)
  • Genetics & Genomics (AREA)
  • Biomedical Technology (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a system level testability design multi-objective optimization method based on an improved PBI method, which comprises the steps of initializing a group of uniformly distributed reference vectors, calculating a penalty factor for each reference vector, iteratively searching an optimal influence factor vector based on a genetic algorithm, increasing the penalty factor in the searching process, optimally selecting a new population by combining an objective function value and an improved PBI function value, carrying out individual selection operation on the new population, and deleting a dominated solution in a final generation population to obtain a pareto optimal solution set of the influence factor vector. By adopting the method, the convergence effect and the uniformity of the pareto optimal solution of the influence factor vector can be improved while the optimal solution is ensured to be obtained, so that the influence factors are reasonably configured, and the purpose of testability optimal design is achieved.

Description

System-level testability design multi-objective optimization method based on improved PBI method
Technical Field
The invention belongs to the technical field of equipment testability design optimization, and particularly relates to a system-level testability design multi-objective optimization method based on an improved PBI method.
Background
In order to reduce the difficulty of later maintenance of the device, the system should consider testability design in the initial stage of design. Testability refers to the degree to which the state of a system can be accurately detected. In the problem of fault diagnosis for large-scale electronic equipment systems, how to select a test scheme to enable the Fault Detection Rate (FDR), the False Alarm Rate (FAR) and various overhead (time, economy and the like) indexes of testing to simultaneously meet constraint conditions tends to be better, and the method is a problem of continuous exploration in the academic and engineering fields.
In the test optimization problem, the concerned test index has a Fault Detection Rate (FDR), an isolation rate, a False Alarm Rate (FAR), a test Time Cost (TC), a test economic cost (PC), and the like. Increasing system testability means additional test hardware, thus affecting system weight, size, development difficulty, functional impact, and system reliability.
Assuming a total of D influencing factors, x d Is represented by D ═ 1,2, …, D. And normalizing the influence factor value into a variable between 0 and 1, then the influence factor vector X is [ X ═ X 1 ,…,x D ]. Assuming that the number of targets to be optimized is M, the objective function of each optimization target is f m (X)，m＝1,2,…,M。
The test optimization target is to reasonably select and set X (i.e. reasonably develop testability design, reasonably allocate resources and the like) so as to minimize M target functions. In reality, it is generally impossible for M objective functions to reach the optimum simultaneously, so that it is a typical multi-objective optimization problem.
When multiobjective optimization is a minimization optimization problem, it can be expressed by the following formula, i.e. it is necessary to find a suitable X to minimize all M objective functions f (X):
minimizeF(X)＝(f 1 (X),f 2 (X),…,f M (X))
the essential difference from the single-objective optimization problem is that the solutions of the multi-objective optimization problem are not unique, but rather there is a set of optimal solutions consisting of numerous Pareto optimal solutionsAnd (3) combining, wherein each element in the set is called a Pareto optimal solution or a non-poor optimal solution. For the vector F (X) determined by the above formula i ) And F (X) j ) If the two vectors are not equal and F (X) i ) All elements in the solution are not more than F (X) j ) The corresponding position element in (b) is called F (X) i ) Dominating F (X) j )，X j Called the dominant solution, X i Referred to as the non-dominated solution. The set of all non-dominated solutions is called the pareto optimal set.
The current algorithms capable of solving the problems include NSGA-III type algorithm, particle swarm algorithm and the like. The NSGA-III type algorithm is typical, a relatively comprehensive non-dominated solution set can be found, and the operation time of the algorithm is relatively long due to the problems of relatively high time complexity, low convergence rate and the like of dominated relation calculation. The problems of low search speed, high convergence algebra and the like are solved.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide a system-level testability design multi-objective optimization method based on an improved PBI method, which uses an improved PBI selection strategy and individual complementary selection operation to ensure that a pareto optimal solution set of influence factor vectors is obtained and simultaneously improve the convergence effect and the uniformity of the pareto optimal solution of the influence factor vectors, thereby reasonably configuring the influence factors and achieving the aim of testability optimal design.
In order to achieve the above purpose, the system level testability design multi-objective optimization method based on the improved PBI method comprises the following steps:
s1: determining influence factors according to the practical situation of the electronic system, and recording the influence factor vector X ═ X 1 ,…,x D ]Wherein x is d Expressing the normalized value of the D-th influencing factor, wherein D is 1,2, …, D expresses the number of the influencing factors; recording the number of the targets needing to be optimized as M, and determining an objective function f of each optimized target m (X), M is 1,2, …, M, the smaller the objective function value is, the better the combination of influencing factors is;
s2: setting N reference vectors
Figure BDA0002965204620000021
Wherein
Figure BDA0002965204620000022
Represents a reference vector W i The mth element value of (1), 2, …, N;
s3: for each reference vector W i Calculating a tangent value theta of an angle between the tangent value theta and each coordinate axis in the M-dimensional search space i,m Then taking M included angle tangent values theta i,m As a reference vector W i Corresponding penalty factor initial value theta i ；
S4: defining the influencing factor vector X as [ X ] 1 ,…,x N ]As an individual in the genetic algorithm, randomly generating N influence factor vectors in a value space omega of the influence factor vectors to form an initial population P of the genetic algorithm;
s5: judging whether an iteration ending condition of the genetic algorithm is met, if so, ending the iteration, and entering a step S15, otherwise, entering a step S6;
s6: carrying out crossing and mutation operations on individuals in the current population P to generate a sub-population Q;
s7: merging the population P and the population Q and putting the merged population P and the merged population Q into a set S;
s8: respectively calculating an objective function value f of each optimization target corresponding to each individual in the set S j (X k )， k＝1,2,…,2N；
S9: sorting the individuals in the set S in a non-dominated way, and forming the non-dominated individuals which are not dominated by other individuals into the set S nd The rest of the dominated individuals dominated by other individuals form a set S d ；
S10: for each objective function f j (X) from the set S of non-dominant individuals nd Screening each individual for the maximum value of the values of the objective function
Figure BDA0002965204620000031
And minimum value
Figure BDA0002965204620000032
Then according to the followingThe value of the objective function f of the formula for each individual j (X k ) Normalizing to obtain normalized target function value
Figure BDA0002965204620000033
Figure BDA0002965204620000034
S11: preferably obtaining a new population based on an improved PBI method, which comprises the following steps:
s11.1: respectively calculating the PBI function value g (X) of each individual in the set S under each weight vector k |W i ,Z * ) The calculation formula is as follows:
g(X k |W i ,Z * )＝d 1 (k,i)+θ i d 2 (k,i)
wherein, the first and the second end of the pipe are connected with each other,
Figure BDA0002965204620000035
Figure BDA0002965204620000036
s11.2: for each reference vector W i Setting a sub-population phi i For each individual in the set S, the value g (X) of the PBI function is determined from the N PBI functions corresponding to the individual k |W i ,Z * ) Screening out the minimum value, and adding the individual into the sub-population of the reference vector corresponding to the minimum PBI function value;
s11.3: the individuals of the new population are optimized in batches, and each time the individuals are optimized, the individuals are optimized from each sub-population phi i Screening out the individual with minimum function value for corresponding reference vector PBI, adding new population P', and selecting sub-population phi i Deleting the individual, and circulating the process until the number of the individuals in the new population P' is N;
s12: belonging to the non-dominated individual set S in the new population P nd Of (1) constitutes a set P' nd Belonging to a dominant individual set S in the new population P d Of (b) constitutes a set P' d Then set S of non-dominated individuals nd Is not in set P' nd The individuals of (2) constitute a set P * ；
S13: the individual selection is carried out by adopting the following method:
if the set P * Number of individuals | P in * L is less than or equal to set P' d Of (1) | P' d If so, then P * Adding all the individuals in the group A into a complementary selection set add; if set P * Number of individuals | P in * L is greater than set P' d Of (1) | P' d If, then the following method is adopted from the set P * Screening out | P' d L individual constitutes a complement set add:
for set P * Is calculated from the set P' nd Then screening out the individuals with the maximum minimum distance value, adding the individuals into the complementary selection set add, and then selecting the individuals from the set P * Deleting; the process is circulated until the number of individuals in the complement set add is | P' d |；
Adding the complementary selection set add into the population P 'to form a new population P';
s14: let the population P equal to P', penalty factor theta i ＝θ i +1, return to step S5;
s15: and (4) deleting the dominated solution from the population obtained from the algorithm execution to the last generation, wherein the obtained population is the pareto optimal solution set serving as the influence factor vector.
The invention relates to a system-level testability design multi-objective optimization method based on an improved PBI method, which comprises the steps of initializing a group of uniformly distributed reference vectors, calculating a penalty factor for each reference vector, iteratively searching an optimal influence factor vector based on a genetic algorithm, increasing the penalty factor in the searching process, optimally selecting a new population by combining an objective function value and an improved PBI function value, carrying out individual complementary selection operation on the new population, and deleting a dominated solution in a final population to obtain a pareto optimal solution set of the influence factor vectors. By adopting the method, the convergence effect and the uniformity of the pareto optimal solution of the influence factor vector can be improved while the optimal solution is ensured to be obtained, so that the influence factors are reasonably configured, and the purpose of testability optimal design is achieved.
Drawings
FIG. 1 is a schematic diagram of a PBI process;
FIG. 2 is a schematic diagram of penalty factor derivation;
FIG. 3 is a flow chart of an embodiment of the system level testability design multi-objective optimization method based on the improved PBI method of the present invention;
FIG. 4 is a flow chart of the preferred generation of new populations based on the modified PBI method of the present invention;
FIG. 5 is a diagram illustrating an optimal influence factor vector distribution obtained by the present invention;
FIG. 6 is a vector diagram of the optimal influence factor obtained by the MOEA-D algorithm in this embodiment;
FIG. 7 is a vector distribution diagram of the optimal influence factors obtained by using the NSGA-III algorithm in this embodiment;
fig. 8 is a schematic diagram of a standard pareto frontplane model of the present embodiment.
Detailed Description
The following description of the embodiments of the present invention is provided in order to better understand the present invention for those skilled in the art with reference to the accompanying drawings. It is to be expressly noted that in the following description, a detailed description of known functions and designs will be omitted when it may obscure the subject matter of the present invention.
Examples
To better explain the technical solution of the present invention, first, the technical principle of the present invention will be explained.
FIG. 1 is a schematic representation of the PBI process. As shown in fig. 1, in the PBI (penalty-based boundary intersection) method, W is a pre-specified reference vector, which is generally automatically generated according to the number specified by the user, and if a two-dimensional space (one quadrant) is divided into 5, 6 reference vectors are required, and an included angle is formed between the reference vectorsAt 90/5-18. The multi-objective optimization based on this method is to distribute an objective function f (x) on each reference vector and close to the origin of coordinates (minimization problem). Measuring whether an objective function (1) is close to a reference vector; (2) whether it is close to the origin of coordinates can be represented by d in FIG. 1 1 And d 2 Expresses the weighted sum of:
Figure BDA0002965204620000051
wherein the content of the first and second substances,
Figure BDA0002965204620000052
Figure BDA0002965204620000053
Ω denotes an influence factor vector X ═ X 1 ,…,x N ]The value space of (a) is defined,
Figure BDA0002965204620000054
is an ideal point, and the point is that,
Figure BDA0002965204620000055
representing the objective function f of the influencing factor vector X at all points in the value space omega j And (X), superscript T represents transposition, and | | represents norm calculation. FIG. 1 shows F (X) to W (0.5 ) T D of 1 And d 2 。d 1 For measuring whether X converges to the pareto optimum plane, d 2 And whether the solution is close to the reference line or not is measured, so that the solution diversity (whether the solution is uniformly distributed or not) is ensured. g (X | W, Z) * )＝d 1 +θd 2 It can be simultaneously evaluated whether a solution X has both convergence and diversity.
Under the condition that the penalty factor theta is appropriate in value, the PBI method can well solve the problem of any pareto front edge shape. However, the performance of the algorithm is seriously degraded when the PF surface distribution is discontinuous and is not convex, so that the invention further improves the PBI method, deduces and sets a penalty factor and provides a new PBI selection method for optimization; then, the minimum and maximum distance selection is combined to ensure that the algorithm can obtain good effect in the face of any pareto front edge shape. Therefore, the searching work is more effective, the convergence is faster, and the efficiency of the electronic system testability design multi-objective optimization method is improved.
Fig. 2 is a schematic diagram of penalty factor derivation. As shown in fig. 2, the penalty factor is set to select a solution that satisfies the requirement in each reference vector direction. Let the line segment AB be a contour of the reference vector W and AB be perpendicular to the coordinate axis,
Figure BDA0002965204620000061
and alpha is the angle between the reference vector W and the coordinate axis. Then there are:
Figure BDA0002965204620000062
and because:
Figure BDA0002965204620000063
therefore, the following can be obtained:
Figure BDA0002965204620000064
due to the fact that
Figure BDA0002965204620000065
Therefore, it is possible to
Figure BDA0002965204620000066
That is, theta is the tangent of the reference vector to the coordinate axis.
For M optimization targets, the search space is an M-dimensional space, each reference vector W and M coordinate axes form M included angles, and tangent values of the M included angles are obtainedTo obtain theta W ＝(θ 1 ,…,θ M ). From the analysis of the contour lines, the largest one is taken as the lower bound of the penalty value of this reference vector, i.e.θ＝max(θ 1 ,…,θ M ). From the above analysis, the setting range of the penalty factor is θ ∈ [ [ alpha ], [ alpha ] ]θInfinity), where the lower boundθIs the penalty value corresponding to the boundary of the dominant region with the smallest angle with the reference vector W.
Based on the analysis, the invention provides a system level testability design multi-objective optimization method based on an improved PBI method. FIG. 3 is a flow chart of an embodiment of the system level testability design multi-objective optimization method based on the improved PBI method of the present invention. As shown in fig. 3, the system-level testability design multi-objective optimization method based on the improved PBI method of the present invention specifically includes the following steps:
s301: determining influence factors and optimizing an objective function:
determining influence factors according to the practical condition of the electronic system, and recording the vector X of the influence factors as [ X ] 1 ,…,x D ]Wherein x is d Expressing the normalized value of the D-th influencing factor, wherein D is 1,2, …, D expresses the number of the influencing factors; recording the number of the targets needing to be optimized as M, and determining an objective function f of each optimized target m (X), M is 1,2, …, M, and the smaller the objective function value, the better the combination of influencing factors.
S302: generating a reference vector:
setting N reference vectors
Figure BDA0002965204620000067
Wherein
Figure BDA0002965204620000068
Represents a reference vector W i I-1, 2, …, N.
In this embodiment, the reference vectors are generated by simplex method, and N reference vectors W i Are uniformly distributed. The number of reference vectors N can be calculated using the following formula:
Figure BDA0002965204620000071
wherein H represents a preset constant parameter.
A multi-objective problem can be decomposed into N sub-problems using N reference vectors, one reference vector corresponding to each sub-problem, and optimization of the multi-objective optimization problem is accomplished by optimizing each sub-problem (optimization in the direction of each reference vector).
S303: initializing a penalty factor:
for each reference vector W i Calculating a tangent theta of an angle formed between the tangent and each coordinate axis in the M-dimensional search space i,m Then taking M included angle tangent values theta i,m As the reference vector W i Corresponding penalty factor initial value theta i 。
In the subsequent iteration process, as the population is gradually converged, the maintenance of the diversity of the population is critical at this time, and therefore a penalty factor is set to be increased progressively according to the iteration times.
S304: initializing a population:
defining the influencing factor vector X as [ X ] 1 ,…,x N ]As an individual in the genetic algorithm, N influence factor vectors are randomly generated in the value space omega of the influence factor vectors to form an initial population P of the genetic algorithm.
S305: whether an iteration end condition is reached:
and judging whether an iteration ending condition of the genetic algorithm is reached, if so, ending the iteration, and entering the step S315, otherwise, entering the step S306. The iteration ending conditions of the genetic algorithm are generally two, one is the maximum iteration number, the other is an objective function threshold, and one is selected according to actual needs.
S306: generating a sub-population:
and carrying out cross and variation operation on the individuals in the current population P to generate a sub-population Q. In the present embodiment, a classical SBX crossover algorithm and a polynomial mutation algorithm are used.
S307: merging the populations:
and merging the population P and the population Q into a set S, wherein the number of individuals in the set S is 2N obviously.
S308: calculating an objective function value:
respectively calculating an objective function value f of each optimization target corresponding to each individual in the set S j (X k )， k＝1,2,…,2N。
S309: non-dominant ordering:
sorting the individuals in the set S in a non-dominated way, and forming the non-dominated individuals which are not dominated by other individuals into the set S nd The other dominated individuals dominated by other individuals constitute a set S d 。
S310: normalizing the objective function value:
for each objective function f j (X) from the set S of non-dominated individuals nd For each individual, selecting the maximum value from the values of the objective function
Figure BDA0002965204620000081
And minimum value
Figure BDA0002965204620000082
The objective function value f for each individual is then calculated according to the following formula j (X k ) Normalizing to obtain normalized target function value
Figure BDA0002965204620000083
Figure BDA0002965204620000084
Thus, a normalized objective function vector for each individual can be obtained
Figure BDA0002965204620000085
Using a set S of non-dominated individuals nd The maximum value and the minimum value of the determined objective function value are more reasonable, and the iteration efficiency is improved.
S311: preferably, a new population is obtained based on the improved PBI method:
FIG. 4 is a flow chart of the preferred generation of new populations based on the improved PBI method of the present invention. As shown in fig. 4, the specific steps of the present invention for preferably obtaining a new population based on the improved PBI method include:
s401: calculating PBI function value:
the PBI function value g (X) of each individual in the set S is calculated under each reference vector k |W i ,Z * ) Since the normalization of the objective function value has been performed in step S310, the calculation formula of the PBI function value is as follows:
g(X k |W i ,Z * )＝d 1 (k,i)+θ i d 2 (k,i) (9)
wherein the content of the first and second substances,
Figure BDA0002965204620000086
Figure BDA0002965204620000087
s402: and (3) dividing the population:
for each reference vector W i Setting a sub-population phi i For each individual in the set S, the value g (X) of the PBI function is calculated from the corresponding N PBI functions k |W i ,Z * ) And (5) screening out the minimum value, and adding the individual into the sub population of the reference vector corresponding to the minimum PBI function value. Obviously, in N sub-populations φ i Some of the sub-populations are empty, some of the sub-populations comprise one individual, and some of the sub-populations comprise a plurality of individuals.
S403: preferably a new population of individuals:
next, a new population of individuals is preferred in batches, each time an individual is preferred, from each sub-population phi i Selecting the individual with the minimum function value of the corresponding reference vector PBI, adding a new population P', and selecting a sub-population phi i The process is cycled through until the number of individuals in the new population P' is N.
S312: and (3) carrying out classification marking on new population individuals:
belonging to the non-dominated individual set S in the new population P nd Of (b) constitutes a set P' nd Belonging to a dominant individual set S in the new population P d Of (1) constitutes a set P' d Then set S of non-dominated individuals nd Does not belong to set P' nd Form a set P * 。
S313: and (3) individual additional selection:
since the set S of non-dominant individuals has already been set S in step S312 nd In the group of individuals not selected by the new population form a set P * In order to make the number of non-dominant individuals in the new population as much as possible, the following method can be adopted for individual selection:
if the set P * Number of individuals | P in * L is less than or equal to set P' d Of (1) | P' d If P is not equal to P * All of the individuals in the group add to the complementary selection set add. If set P * Number of individuals | P in * L is greater than set P' d Of (1) | P' d If, then the following method is adopted from the set P * Screening out | P' d L individual constitutes a complement set add:
for set P * Is calculated with the set P' nd Of (i.e. and set P' nd The minimum value of the distances of all individuals), then screening out the individual with the maximum minimum distance value, adding the individual into the complementary selection set add, and then selecting the individual from the set P * Is deleted. The process is circulated until the number of individuals in the complement set add is | P' d |。
And adding the complementary selection set add into the population P 'to form a new population P'. I.e. P ═ P @ uadd.
It is worth noting that during evolution, individuals within a population will change, but the total number will not exceed 2N.
By this way of individual class labeling and reselection, the number of dominant solutions can be made as large as possible while preserving the dominant solution, which has two benefits:
(1) because the dominated solution holds the search information in the direction of its reference vector, by retaining the dominated solution during the evolution process, it is guaranteed that each reference vector always maintains the search in its direction. So that solutions for certain regions are not lost.
(2) For some problems, such as degradation problems, the PBI selection operation may find less non-dominant solutions. In order to better cover PF, the non-dominant individual obtained by the complementary selection is added into the population through the individual complementary selection operation, so that the number of non-dominant solutions in the population can reach the population size N, and the requirement of diversity is met.
S314: let the population P equal to P', penalty factor theta i ＝θ i +1, return to step S305.
S315: obtaining a pareto optimal solution set:
and (4) deleting the dominated solution from the population obtained from the algorithm execution to the last generation, wherein the obtained population is the pareto optimal solution set serving as the influence factor vector.
Examples
In order to better explain the technical scheme of the invention, the following takes three-object optimization as an example to explain the specific implementation process of the invention. The optimized target of the testability design of the anti-tank missile launching system is assumed to be the maximum fault detection rate FDR, and the expression is f 1 Max mize (fdr); minimum false alarm rate FAR, expression f 2 Minimize (far); and a test cost C, expressed as f 3 Minimize (c). Let f 1 1-maxmize (fdr), all translate to minimization problems. There are many factors that affect the three targets, such as design difficulty, volume consideration, functional influence, reliability influence, and the like, and in this embodiment, 7 influencing factors are selected, including test correlation coefficient, test omission probability, test false alarm probability, failure prior probability, misdiagnosis cost, missed diagnosis cost, and test cost, that is, X ═ X 1 ,…,x 7 ]。
The objective function F ═ F constructed in this embodiment 1 ,f 2 ,f 3 ]And the optimization problem is as follows:
Minimize f q ＝a q ×(1+h(X)),q＝1,2,3 (12)
wherein:
Figure BDA0002965204620000101
Figure BDA0002965204620000102
in this embodiment, since there are 3 optimization targets, the objective function space is a three-dimensional space, and each bit of the three-dimensional space is equally divided into 12 parts, so that there are 3 optimization targets in total
Figure BDA0002965204620000103
A reference point. The population number is 91 and the maximum iteration number G is set max 1000. Table 1 shows objective function vectors corresponding to the optimal solutions obtained by the present invention in this embodiment. Table 2 is the influencing factor vector corresponding to the optimal objective function vector in table 1.
Figure BDA0002965204620000104
Figure BDA0002965204620000111
Figure BDA0002965204620000121
Figure BDA0002965204620000131
TABLE 1
Figure BDA0002965204620000132
Figure BDA0002965204620000141
Figure BDA0002965204620000151
TABLE 2
To illustrate the technical effect of the present invention, the present example (number of individuals 91, generation 1000) was run using the currently very widely used MOEA-D algorithm and NSGA-III algorithm, and the results were compared with those of the present invention. In order to more vividly illustrate the simulation results, the results are shown in a 3-dimensional perspective view. Fig. 5 is a distribution diagram of the optimal influence factor vectors obtained by using the present invention in this embodiment. Fig. 6 is a distribution diagram of the optimal influence factor vector obtained by using the MOEA-D algorithm in the present embodiment. FIG. 7 is the vector distribution diagram of the optimal influence factors obtained by using the NSGA-III algorithm in this embodiment. Fig. 8 is a schematic diagram of a standard pareto frontplane model of the present embodiment. Comparing fig. 5 to fig. 8, it can be seen that although some test points can be obtained by using MOEA-D and NSGA-III algorithms, the distribution is not uniform, and a sufficient number of optimal test points cannot be found.
The Inverse Generation Distance (IGD) is a comprehensive performance evaluation index, and the convergence performance and the distribution performance of the method are evaluated by calculating the minimum Distance sum between each point (individual) on the real pareto front surface and the set of individuals obtained by the algorithm. The smaller the value of the anti-generation distance, the better the overall performance of the method, including convergence and distribution performance. The IGD of the invention is 0.05704, which is better than 0.74320 of MOEA-D algorithm and 0.38704 of NSGA-III algorithm.
In conclusion, the invention can find a more uniformly distributed and better test scheme when diagnosing the fault of the anti-tank missile transmitting system. The testability designer can reasonably configure influence factors according to the requirement importance of the three functions (detection rate, false alarm rate and fault diagnosis cost) in different occasions and the operation result, so that the aim of testability optimal design is fulfilled.
Although illustrative embodiments of the present invention have been described above to facilitate the understanding of the present invention by those skilled in the art, it should be understood that the present invention is not limited to the scope of the embodiments, and various changes may be made apparent to those skilled in the art as long as they are within the spirit and scope of the present invention as defined and defined by the appended claims, and all matters of the invention which utilize the inventive concepts are protected.

Claims (2)

1. A system-level testability design multi-objective optimization method based on an improved PBI method is characterized by comprising the following steps:
s1: determining influence factors according to the actual conditions of the anti-tank missile launching system, wherein the influence factors comprise a test correlation coefficient, a test omission probability, a test false alarm probability, a fault prior probability, a misdiagnosis cost, a missed diagnosis cost and a test cost, and recording an influence factor vector X ═ X 1 ,…,x D ]Wherein x is d Expressing the normalized value of the D-th influencing factor, wherein D is 1,2, …, D expresses the number of the influencing factors; the optimization targets comprise a maximum fault detection rate FDR, a minimum false alarm rate FAR and a test cost C, the number of the targets needing to be optimized is recorded as M, and an objective function f of each optimization target is determined m (X), M is 1,2, …, M, the smaller the objective function value is, the better the combination of influencing factors is;
s2: setting N reference vectors
Figure FDA0003640436720000011
Wherein
Figure FDA0003640436720000012
Represents a reference vector W i The mth element value of (1), 2, …, N;
s3: for each reference vector W i Calculating a tangent theta of an angle formed between the tangent and each coordinate axis in the M-dimensional search space i,m Then taking M included angle tangent values theta i,m As a reference vector W i Corresponding penalty factor initial value theta i ；
S4: defining the influencing factor vector X as [ X ] 1 ,…,x N ]As an individual in the genetic algorithm, randomly generating N influence factor vectors in a value space omega of the influence factor vectors to form an initial population P of the genetic algorithm;
s5: judging whether an iteration end condition of the genetic algorithm is reached, if so, ending the iteration, and entering the step S15, otherwise, entering the step S6;
s6: carrying out cross and variation operation on individuals in the current population P to generate a sub-population Q;
s7: merging the population P and the population Q and putting the population P and the population Q into a set S;
s8: respectively calculating an objective function value f of each optimization target corresponding to each individual in the set S j (X k )，k＝1,2,…,2N；
S9: sorting the individuals in the set S in a non-dominated way, and forming the non-dominated individuals which are not dominated by other individuals into the set S nd The rest of the dominated individuals dominated by other individuals form a set S d ；
S10: for each objective function f j (X) from the set S of non-dominant individuals nd For each individual, selecting the maximum value from the values of the objective function
Figure FDA0003640436720000013
And minimum value
Figure FDA0003640436720000014
The objective function value f for each individual is then calculated according to the following formula j (X k ) Normalizing to obtain normalized target function value
Figure FDA0003640436720000015
Figure FDA0003640436720000016
S11: preferably obtaining a new population based on an improved PBI method, which comprises the following steps:
s11.1: respectively calculating the PBI function value g (X) of each individual in the set S under each weight vector k |W i ,Z * ) The calculation formula is as follows:
g(X k |W i ,Z * )＝d 1 (k,i)+θ i d 2 (k,i)
wherein the content of the first and second substances,
Figure FDA0003640436720000021
Figure FDA0003640436720000022
Figure FDA0003640436720000023
s11.2: for each reference vector W i Setting a sub-population phi i For each individual in the set S, the value g (X) of the PBI function is determined from the N PBI functions corresponding to the individual k |W i ,Z * ) Screening out the minimum value, and adding the individual into a sub-population of the reference vector corresponding to the minimum PBI function value;
s11.3: selecting new population of individuals in batches, and selecting individuals from each sub-population phi each time i Screening out the individual with minimum function value for corresponding reference vector PBI, adding new population P', and selecting sub-population phi i Deleting the individual, and circulating the process until the number of the individuals in the new population P' is N;
s12: the non-dominant individual in the new population P' is collected S nd Of (b) constitutes a set P' nd Belonging to a dominant individual set S in the new population P d Of (1) constitutes a set P' d Then set S of non-dominant individuals nd In not belonging to the setP′ nd The individuals of (2) constitute a set P * ；
S13: the individual selection is carried out by adopting the following method:
if set P * Number of individuals | P in * Is less than or equal to the set P d 'Individual number of | P' d If P is not equal to P * All of the individuals in the group add to the complementary selection set add. If the set P * Number of individuals | P in * L is greater than set P' d Of (2) | P' d If, then the following method is adopted from the set P * Screening out | P' d L individual constitutes a complement set add:
for set P * Is calculated with the set P n ′ d Then screening out the individuals with the maximum minimum distance value, adding the individuals into the complementary selection set add, and then selecting the individuals from the set P * Deleting; the process is circulated until the number of individuals in the complement set add is | P' d |；
Adding the complementary selection set add into the population P 'to form a new population P';
s14: let the population P equal to P', penalty factor theta i ＝θ i +1, return to step S5;
s15: and (4) deleting the dominated solution from the population obtained by the algorithm execution to the last generation, wherein the obtained population is the pareto optimal solution set serving as the influence factor vector.
2. The system-level design-for-test multi-objective optimization method of claim 1, wherein the number N of reference vectors in step S2 is calculated by using the following formula:
Figure FDA0003640436720000031
wherein H represents a preset constant parameter.
CN202110249020.2A 2021-03-08 2021-03-08 System-level testability design multi-objective optimization method based on improved PBI method Expired - Fee Related CN112883651B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110249020.2A CN112883651B (en) 2021-03-08 2021-03-08 System-level testability design multi-objective optimization method based on improved PBI method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110249020.2A CN112883651B (en) 2021-03-08 2021-03-08 System-level testability design multi-objective optimization method based on improved PBI method

Publications (2)

Publication Number Publication Date
CN112883651A CN112883651A (en) 2021-06-01
CN112883651B true CN112883651B (en) 2022-07-26

Family

ID=76055686

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110249020.2A Expired - Fee Related CN112883651B (en) 2021-03-08 2021-03-08 System-level testability design multi-objective optimization method based on improved PBI method

Country Status (1)

Country Link
CN (1) CN112883651B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113376593B (en) * 2021-06-10 2022-10-04 西安电子科技大学 STAP radar distributed interference method based on MPBI decomposition
CN117279009A (en) * 2023-09-23 2023-12-22 北京理工大学 Network slice reconstruction method based on countermeasure generation network assisted pareto optimization
CN117829001B (en) * 2024-03-05 2024-05-31 四川汇利实业有限公司 Genetic algorithm-based sizing amount control method, system, equipment and medium

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109597758A (en) * 2018-11-07 2019-04-09 电子科技大学 System level testing design optimization method based on PBI
CN109858093A (en) * 2018-12-28 2019-06-07 浙江工业大学 The air source heat pump multi-objective optimization design of power method of the non-dominated sorted genetic algorithm of SVR neural network aiding
CN110908897A (en) * 2019-10-29 2020-03-24 电子科技大学 System-level testability design multi-objective optimization method based on self-adaptive search direction

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7437336B2 (en) * 2003-08-01 2008-10-14 George Mason Intellectual Properties, Inc. Polyoptimizing genetic algorithm for finding multiple solutions to problems

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109597758A (en) * 2018-11-07 2019-04-09 电子科技大学 System level testing design optimization method based on PBI
CN109858093A (en) * 2018-12-28 2019-06-07 浙江工业大学 The air source heat pump multi-objective optimization design of power method of the non-dominated sorted genetic algorithm of SVR neural network aiding
CN110908897A (en) * 2019-10-29 2020-03-24 电子科技大学 System-level testability design multi-objective optimization method based on self-adaptive search direction

Non-Patent Citations (8)

* Cited by examiner, † Cited by third party
Title
A modified PBI approach for multi-objective optimization with complex pareto fronts;Qisheng zhang等;《Swarm and Evolutioary Computation》;20180630;第40卷;全文 *
A vector angle-based evolutuionary algorithm for unocnstrained many-objective optimization;Yi Xiang等;《IEEE transactions on evolutionary computation》;20170228;第21卷(第1期);全文 *
PBI function based evolutionary algorithm with precise penalty parameter for unconstrained many-objective optimization;yang chenglin等;《Swarm and Evolutioary Computation》;20191130;第50卷;全文 *
一种用于多目标约束优化的改进进化算法;俞国燕等;《计算机集成制造系统》;20090615(第06期);1173-1178页 *
动态惩罚分解策略下的高维目标进化算法;王丽萍等;《小型微型计算机系统》;20181015(第10期);28-35页 *
基于PBI函数的多目标进化算法研究;邹宇;《中国优秀硕士学位论文全文数据库 信息科技辑》;20200715(第7期);I140-145 *
基于角度惩罚距离精英选择策略的偏好高维目标优化算法;王丽萍等;《计算机学报》;20180131(第01期);238-255页 *
改进的约束优化多目标遗传算法及工程应用;王俊年等;《计算机工程与应用》;20061231(第03期);247-253页 *

Also Published As

Publication number Publication date
CN112883651A (en) 2021-06-01

Similar Documents

Publication Publication Date Title
CN112883651B (en) System-level testability design multi-objective optimization method based on improved PBI method
CN107454105B (en) Multidimensional network security assessment method based on AHP and grey correlation
Boulkaibet et al. Finite element model updating using fish school search and volitive particle swarm optimization
Zhou et al. A surrogate-assisted evolutionary algorithm for minimax optimization
CN107909194B (en) System-level testability design multi-objective optimization method
CN109597758B (en) System-level testability design optimization method based on PBI
CN106060008B (en) A kind of network intrusions method for detecting abnormality
CN110768971B (en) Confrontation sample rapid early warning method and system suitable for artificial intelligence system
CN110908897B (en) System-level testability design multi-objective optimization method based on self-adaptive search direction
CN112632765B (en) Combat capability assessment method combining weighting method and SEM method
CN104850867A (en) Object identification method based on intuitive fuzzy c-means clustering
Wu et al. A low-sample-count, high-precision Pareto front adaptive sampling algorithm based on multi-criteria and Voronoi
Shirk et al. The effect of gene flow from unsampled demes in landscape genetic analysis
CN115112372A (en) Bearing fault diagnosis method and device, electronic equipment and storage medium
Amrit et al. Design strategies for multi-objective optimization of aerodynamic surfaces
CN116738920B (en) Chip 3D design method and system of three-dimensional geometric kernel
KR101462349B1 (en) Method for multi-objective optimizing based on Pareto-optimality
Liao et al. A novel classification and identification scheme of emitter signals based on ward’s clustering and probabilistic neural networks with correlation analysis
Guo et al. Data mining and application of ship impact spectrum acceleration based on PNN neural network
CN108830432B (en) Unmanned aerial vehicle group action scheme searching method based on small amount of prior knowledge
CN116523640A (en) Financial information management system based on scheduling feedback algorithm
CN114092472B (en) Method, device and medium for detecting uncertain samples in defect detection
Wang et al. FCM algorithm and index CS for the signal sorting of radiant points
Fan et al. Efficient constrained large-scale multi-objective optimization based on reference vector-guided evolutionary algorithm
CN113239469B (en) Structure optimization method, device, equipment and storage medium for vehicle body parts

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20220726